IEEE Geoscience and Remote Sensing Magazine - March 2020 - 35

LARGE-SCENE PROCESSING
Sometimes large-scale deformation maps must be processed, such as the deformation retrieval of a whole city.
The German Aerospace Center calls processing large-scene
images using the time-series InSAR technique, wide area
product (WAP) processing [118]-[120]. The goal of WAP processing is to estimate the deformation of several thousand
square kilometers or more. Initially, it was mainly aimed at
the Terrain Observation by Progressive Scan imaging mode
of Sentinel-1. However, there are still some problems in the
processing of large scenes. First, as the scene increases, the
image contains more and more complex objects. The same
area may include many ground scenes, such as urban buildings, mountainous areas, farmland, rivers, and lakes. The
complex ground environment leads to the uneven distribution of PSs and DSs, and their densities vary spatially.
This makes it difficult to establish the PS network and solve
the arc parameters, which may bring about spatial-error
propagation. Second, with the increased area size, some approximate relationships are no longer valid. The most obvious thing is to consider the changes of spatial baseline
with range and azimuth direction. Third, large-scale processing is inevitably accompanied by a large amount of data
that have high requirements on the hardware platform's

computing performance and storage space. It is also necessary to optimize and accelerate the algorithm.
To solve this problem, a series of refined processing
methods are proposed. The focus is on establishing the PS
network and estimating the arc parameters. In [121], the
least squares ambiguity decorrelation adjustment (LAMBDA) method is used to solve the arc parameters of the network first. Then the outliers are identified and eliminated
using the L1-norm-based network inversion technique.
The remaining network is inverted under the L2-norm
to retrieve absolute parameters. In [123], the authors improved the construction of the reference network. First, the
PSs in the scene were classified by the K-means clustering
method. Then, the PS network was formed based on the
maximum range and maximum number of arcs per cluster. The parameters of the arc were estimated; repetitive
arcs and arcs with poor quality were discarded. Finally, the
relative parameters of the network were integrated via a
least squares inversion.
In addition, for the high time and space complexity of
the algorithm in large-scale phase unwrapping and the discontinuity in blocked processing, there are several methods
that offer solutions. Based on the statistical-cost, networkflow phase-unwrapping algorithm, the authors in [123]

N
A
B

Deformation Rate (mm/Year)
-60

(b)
10

-10

Deformation (mm)

(a)
10

-20
-30
-40
-50
-60

-30
-40
-50
-60
-70

-70
-80
0

-20

100

150

200 250 300
Time (Days)
(c)

350

400

450

-80
0

100

150

200 250 300
Time (Days)
(d)

350

400

450

FIGURE 16. The Danba deformation obtained from (a) PSI and (b) JSInSAR. The time-series deformation of (c) point A and (d) point B.
MARCH 2020

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

IEEE Geoscience and Remote Sensing Magazine - March 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2020